DESCRIPTION (Applicant's abstract): It is important that mental health care policies are based on reliable inferences about medical care costs and reliable predictions about future costs for medical care. However, three characteristics of medical cost data can impede reliable inferences and prediction: (1) highly skewed non-zero cost observations, (2) a certain proportion of zero costs, (3) and incomplete follow-up data Without taking these three characteristics into account, statistical analyses of cost data can lead to unreliable inferences and predictions. In this proposal, we will develop a general methodology for comparisons of means of two or more groups and for regression analysis of cost data, adjusting for these three characteristics.